servediff 0.3.0__tar.gz → 0.3.2__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {servediff-0.3.0 → servediff-0.3.2}/CHANGELOG.md +14 -0
- {servediff-0.3.0 → servediff-0.3.2}/PKG-INFO +7 -3
- {servediff-0.3.0 → servediff-0.3.2}/README.md +6 -2
- {servediff-0.3.0 → servediff-0.3.2}/pyproject.toml +1 -1
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff/cli.py +16 -1
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/PKG-INFO +7 -3
- {servediff-0.3.0 → servediff-0.3.2}/tests/test_cli.py +27 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/test_env.py +12 -0
- {servediff-0.3.0 → servediff-0.3.2}/CONTRIBUTING.md +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/LICENSE +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/MANIFEST.in +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/README.zh-CN.md +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/examples/issue-excerpts/vllm-47047-v0.23-default.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/examples/issue-excerpts/vllm-47047-v0.23-marlin.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/examples/issue-excerpts/vllm-53472-v0.23.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/examples/issue-excerpts/vllm-53472-v0.26-flag.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/examples/issue-excerpts/vllm-53472-v0.26.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/setup.cfg +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff/__init__.py +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff/__main__.py +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff/env.py +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/SOURCES.txt +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/dependency_links.txt +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/entry_points.txt +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/requires.txt +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/src/servediff.egg-info/top_level.txt +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/README.md +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/vllm-47047-v0.23-default.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/vllm-47047-v0.23-marlin.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/vllm-53472-v0.23.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/vllm-53472-v0.26-flag.log +0 -0
- {servediff-0.3.0 → servediff-0.3.2}/tests/fixtures/vllm-53472-v0.26.log +0 -0
|
@@ -1,5 +1,19 @@
|
|
|
1
1
|
# Changelog
|
|
2
2
|
|
|
3
|
+
## 0.3.2 — 2026-11-02
|
|
4
|
+
|
|
5
|
+
- Extract the actual parallel configuration (`parallel.world_size` and the
|
|
6
|
+
DP/PP/PCP/TP/ETP/EP/EPLB role assignment) and the KV cache layout
|
|
7
|
+
(`Using LBNHC KV cache layout`, 0.30+). Found by dogfooding the parser against
|
|
8
|
+
full vLLM 0.28/0.30 startup logs; the role-assignment line is the only place
|
|
9
|
+
TP>1 and expert parallel appear in the log, so it makes TP4+EP configs visible
|
|
10
|
+
to `servediff diff`.
|
|
11
|
+
|
|
12
|
+
## 0.3.1 — 2026-10-31
|
|
13
|
+
|
|
14
|
+
- `--markdown` now also works on `servediff env` snapshots, so a `capture`-based
|
|
15
|
+
environment diff can be rendered as a paste-ready report.
|
|
16
|
+
|
|
3
17
|
## 0.3.0 — 2026-10-31
|
|
4
18
|
|
|
5
19
|
- `servediff capture -o run.log -- <server command>`: run a server, tee its
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: servediff
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.2
|
|
4
4
|
Summary: Diff what your vLLM / SGLang server actually ran with: auto-selected backends, CUDA graphs, startup failures, dependency versions.
|
|
5
5
|
License-Expression: Apache-2.0
|
|
6
6
|
Project-URL: Homepage, https://github.com/apex-mochen/servediff
|
|
@@ -44,8 +44,8 @@ builds. Today you find that by reading startup logs and bisecting by hand.
|
|
|
44
44
|
## Try it in one minute (no GPU)
|
|
45
45
|
|
|
46
46
|
```bash
|
|
47
|
+
pip install servediff
|
|
47
48
|
git clone https://github.com/apex-mochen/servediff && cd servediff
|
|
48
|
-
pip install .
|
|
49
49
|
servediff diff examples/issue-excerpts/vllm-47047-v0.23-default.log examples/issue-excerpts/vllm-47047-v0.23-marlin.log
|
|
50
50
|
```
|
|
51
51
|
|
|
@@ -149,6 +149,8 @@ contains `TOKEN`, `KEY`, `SECRET`, `PASSWORD`, `CREDENTIAL` or `AUTH` are replac
|
|
|
149
149
|
| MoE backend | three log formats seen in 0.23–0.30 | `moe_runner_backend` arg |
|
|
150
150
|
| CUDA graph | eager mode, `cudagraph_mode`, capture sets, pool size | prefill CUDA graph disabled |
|
|
151
151
|
| model runner, UVA fallback | `Using V2 Model Runner`, pinned-memory fallback | — |
|
|
152
|
+
| parallel config | `world_size`, DP/PP/PCP/TP/ETP/EP/EPLB role assignment | — |
|
|
153
|
+
| KV cache layout | `Using LBNHC KV cache layout` (0.30+) | — |
|
|
152
154
|
| startup outcome | ready / failed + root-cause error | ready |
|
|
153
155
|
|
|
154
156
|
Every fact keeps the log line it came from (`servediff show <log>` or `--json`).
|
|
@@ -186,7 +188,9 @@ Exit codes: `0` no differences, `1` differences, `2` nothing recognised.
|
|
|
186
188
|
|
|
187
189
|
- Tested on: log excerpts posted in vllm#47047 (v0.23) and vllm#53472 (v0.23, v0.26);
|
|
188
190
|
full startup logs of vLLM 0.28.0 and 0.30.0 and one SGLang log, all single-GPU with a
|
|
189
|
-
dense model.
|
|
191
|
+
dense model. The parallel-config rules were written against the v0.28/v0.30 format
|
|
192
|
+
(incl. TP>1 role assignment) and TP4+EP synthetic logs; TP > 1 real logs, MoE models
|
|
193
|
+
in full logs, and other versions are not yet verified.
|
|
190
194
|
- Parsing is regex-based on log text, which changes between versions. A missed line
|
|
191
195
|
shows up as `<absent>`, never as a wrong value — please report it.
|
|
192
196
|
|
|
@@ -23,8 +23,8 @@ builds. Today you find that by reading startup logs and bisecting by hand.
|
|
|
23
23
|
## Try it in one minute (no GPU)
|
|
24
24
|
|
|
25
25
|
```bash
|
|
26
|
+
pip install servediff
|
|
26
27
|
git clone https://github.com/apex-mochen/servediff && cd servediff
|
|
27
|
-
pip install .
|
|
28
28
|
servediff diff examples/issue-excerpts/vllm-47047-v0.23-default.log examples/issue-excerpts/vllm-47047-v0.23-marlin.log
|
|
29
29
|
```
|
|
30
30
|
|
|
@@ -128,6 +128,8 @@ contains `TOKEN`, `KEY`, `SECRET`, `PASSWORD`, `CREDENTIAL` or `AUTH` are replac
|
|
|
128
128
|
| MoE backend | three log formats seen in 0.23–0.30 | `moe_runner_backend` arg |
|
|
129
129
|
| CUDA graph | eager mode, `cudagraph_mode`, capture sets, pool size | prefill CUDA graph disabled |
|
|
130
130
|
| model runner, UVA fallback | `Using V2 Model Runner`, pinned-memory fallback | — |
|
|
131
|
+
| parallel config | `world_size`, DP/PP/PCP/TP/ETP/EP/EPLB role assignment | — |
|
|
132
|
+
| KV cache layout | `Using LBNHC KV cache layout` (0.30+) | — |
|
|
131
133
|
| startup outcome | ready / failed + root-cause error | ready |
|
|
132
134
|
|
|
133
135
|
Every fact keeps the log line it came from (`servediff show <log>` or `--json`).
|
|
@@ -165,7 +167,9 @@ Exit codes: `0` no differences, `1` differences, `2` nothing recognised.
|
|
|
165
167
|
|
|
166
168
|
- Tested on: log excerpts posted in vllm#47047 (v0.23) and vllm#53472 (v0.23, v0.26);
|
|
167
169
|
full startup logs of vLLM 0.28.0 and 0.30.0 and one SGLang log, all single-GPU with a
|
|
168
|
-
dense model.
|
|
170
|
+
dense model. The parallel-config rules were written against the v0.28/v0.30 format
|
|
171
|
+
(incl. TP>1 role assignment) and TP4+EP synthetic logs; TP > 1 real logs, MoE models
|
|
172
|
+
in full logs, and other versions are not yet verified.
|
|
169
173
|
- Parsing is regex-based on log text, which changes between versions. A missed line
|
|
170
174
|
shows up as `<absent>`, never as a wrong value — please report it.
|
|
171
175
|
|
|
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
|
|
|
4
4
|
|
|
5
5
|
[project]
|
|
6
6
|
name = "servediff"
|
|
7
|
-
version = "0.3.
|
|
7
|
+
version = "0.3.2"
|
|
8
8
|
description = "Diff what your vLLM / SGLang server actually ran with: auto-selected backends, CUDA graphs, startup failures, dependency versions."
|
|
9
9
|
readme = "README.md"
|
|
10
10
|
requires-python = ">=3.10"
|
|
@@ -76,6 +76,13 @@ LINE_RULES: list[tuple[re.Pattern, callable]] = [
|
|
|
76
76
|
(re.compile(r"FlashInfer top-p/top-k sampling (disabled|enabled)"),
|
|
77
77
|
lambda m: [("backend.sampling_flashinfer", m[1])]),
|
|
78
78
|
(re.compile(r"Using (V\d) Model Runner"), lambda m: [("model_runner", m[1])]),
|
|
79
|
+
# Parallel configuration: what TP/EP/PP the process actually runs with. This is
|
|
80
|
+
# the only place TP>1 (and expert parallel) shows up in the startup log.
|
|
81
|
+
(re.compile(r"world_size=(\d+) rank=(\d+) local_rank=(\d+)"),
|
|
82
|
+
lambda m: [("parallel.world_size", int(m[1]))]),
|
|
83
|
+
(re.compile(r"rank \d+ in world size \d+ is assigned as (DP rank \d+, PP rank \d+, PCP rank \d+, TP rank \d+(?:, ETP rank \S+)?, EP rank \S+, EPLB rank \S+)"),
|
|
84
|
+
lambda m: [("parallel.roles", m[1])]),
|
|
85
|
+
(re.compile(r"Using (\S+) KV cache layout"), lambda m: [("kv_cache.layout", m[1])]),
|
|
79
86
|
(re.compile(r"Pinned memory is not available.*falling back to (.+?) for UVA buffers"),
|
|
80
87
|
lambda m: [("uva_buffers", f"fallback: {m[1]}")]),
|
|
81
88
|
# Startup outcome. Without it, facts missing after a crash look like "not printed".
|
|
@@ -349,10 +356,18 @@ def main(argv=None) -> int:
|
|
|
349
356
|
print("diff: give two server logs or two `servediff env` snapshots, not one of each", file=sys.stderr)
|
|
350
357
|
return 2
|
|
351
358
|
rows = envsnap.ranked(envsnap.diff(json.loads(text_a), json.loads(text_b)))
|
|
359
|
+
key = [r for r in rows if envsnap.is_key(r)]
|
|
352
360
|
if args.json:
|
|
353
361
|
print(json.dumps({"differences": len(rows), "rows": rows}, indent=2, ensure_ascii=False))
|
|
362
|
+
elif args.markdown:
|
|
363
|
+
shown = rows if args.all else key
|
|
364
|
+
print(_render_markdown(
|
|
365
|
+
path_a=str(args.baseline), path_b=str(args.candidate),
|
|
366
|
+
engine_a=None, engine_b=None,
|
|
367
|
+
rows=shown, warnings=[],
|
|
368
|
+
differing=len(key), identical=len(rows) - len(key),
|
|
369
|
+
all_rows=bool(args.all)))
|
|
354
370
|
else:
|
|
355
|
-
key = [r for r in rows if envsnap.is_key(r)]
|
|
356
371
|
shown = rows if args.all else key
|
|
357
372
|
print(f"{'fact':32s} {'baseline':40s} {'candidate':40s} status")
|
|
358
373
|
for r in shown:
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
Metadata-Version: 2.4
|
|
2
2
|
Name: servediff
|
|
3
|
-
Version: 0.3.
|
|
3
|
+
Version: 0.3.2
|
|
4
4
|
Summary: Diff what your vLLM / SGLang server actually ran with: auto-selected backends, CUDA graphs, startup failures, dependency versions.
|
|
5
5
|
License-Expression: Apache-2.0
|
|
6
6
|
Project-URL: Homepage, https://github.com/apex-mochen/servediff
|
|
@@ -44,8 +44,8 @@ builds. Today you find that by reading startup logs and bisecting by hand.
|
|
|
44
44
|
## Try it in one minute (no GPU)
|
|
45
45
|
|
|
46
46
|
```bash
|
|
47
|
+
pip install servediff
|
|
47
48
|
git clone https://github.com/apex-mochen/servediff && cd servediff
|
|
48
|
-
pip install .
|
|
49
49
|
servediff diff examples/issue-excerpts/vllm-47047-v0.23-default.log examples/issue-excerpts/vllm-47047-v0.23-marlin.log
|
|
50
50
|
```
|
|
51
51
|
|
|
@@ -149,6 +149,8 @@ contains `TOKEN`, `KEY`, `SECRET`, `PASSWORD`, `CREDENTIAL` or `AUTH` are replac
|
|
|
149
149
|
| MoE backend | three log formats seen in 0.23–0.30 | `moe_runner_backend` arg |
|
|
150
150
|
| CUDA graph | eager mode, `cudagraph_mode`, capture sets, pool size | prefill CUDA graph disabled |
|
|
151
151
|
| model runner, UVA fallback | `Using V2 Model Runner`, pinned-memory fallback | — |
|
|
152
|
+
| parallel config | `world_size`, DP/PP/PCP/TP/ETP/EP/EPLB role assignment | — |
|
|
153
|
+
| KV cache layout | `Using LBNHC KV cache layout` (0.30+) | — |
|
|
152
154
|
| startup outcome | ready / failed + root-cause error | ready |
|
|
153
155
|
|
|
154
156
|
Every fact keeps the log line it came from (`servediff show <log>` or `--json`).
|
|
@@ -186,7 +188,9 @@ Exit codes: `0` no differences, `1` differences, `2` nothing recognised.
|
|
|
186
188
|
|
|
187
189
|
- Tested on: log excerpts posted in vllm#47047 (v0.23) and vllm#53472 (v0.23, v0.26);
|
|
188
190
|
full startup logs of vLLM 0.28.0 and 0.30.0 and one SGLang log, all single-GPU with a
|
|
189
|
-
dense model.
|
|
191
|
+
dense model. The parallel-config rules were written against the v0.28/v0.30 format
|
|
192
|
+
(incl. TP>1 role assignment) and TP4+EP synthetic logs; TP > 1 real logs, MoE models
|
|
193
|
+
in full logs, and other versions are not yet verified.
|
|
190
194
|
- Parsing is regex-based on log text, which changes between versions. A missed line
|
|
191
195
|
shows up as `<absent>`, never as a wrong value — please report it.
|
|
192
196
|
|
|
@@ -197,3 +197,30 @@ def test_capture_reports_missing_command(tmp_path, capsys):
|
|
|
197
197
|
log = tmp_path / "x.log"
|
|
198
198
|
assert servediff.main(["capture", "-o", str(log), "--", "no-such-binary-xyz"]) == 2
|
|
199
199
|
assert "cannot start" in capsys.readouterr().err
|
|
200
|
+
|
|
201
|
+
|
|
202
|
+
TP4_EP = """\
|
|
203
|
+
(EngineCore pid=100) INFO 10-28 11:28:12 [parallel_state.py:1827] world_size=4 rank=2 local_rank=2 distributed_init_method=file:///tmp/vllm_dist_x backend=nccl
|
|
204
|
+
(EngineCore pid=100) INFO 10-28 11:28:12 [parallel_state.py:2267] rank 2 in world size 4 is assigned as DP rank 0, PP rank 0, PCP rank 0, TP rank 2, ETP rank 2, EP rank 1, EPLB rank 0
|
|
205
|
+
(EngineCore pid=100) INFO 10-28 11:28:34 [utils.py:320] Using LBNHC KV cache layout.
|
|
206
|
+
"""
|
|
207
|
+
|
|
208
|
+
|
|
209
|
+
def test_tp4_ep_parallel_roles_are_parsed():
|
|
210
|
+
f = servediff.parse(TP4_EP)["facts"]
|
|
211
|
+
assert f["parallel.world_size"].value == 4
|
|
212
|
+
assert f["parallel.roles"].value == ("DP rank 0, PP rank 0, PCP rank 0, TP rank 2, "
|
|
213
|
+
"ETP rank 2, EP rank 1, EPLB rank 0")
|
|
214
|
+
assert f["kv_cache.layout"].value == "LBNHC"
|
|
215
|
+
|
|
216
|
+
|
|
217
|
+
def test_roles_without_etp_field_and_kv_layout_absent():
|
|
218
|
+
text = ("(EngineCore pid=3811) INFO 10-28 11:26:32 [parallel_state.py:1638] "
|
|
219
|
+
"world_size=1 rank=0 local_rank=0 distributed_init_method=file:///tmp/vllm_dist_x backend=nccl\n"
|
|
220
|
+
"(EngineCore pid=3811) INFO 10-28 11:26:32 [parallel_state.py:1982] "
|
|
221
|
+
"rank 0 in world size 1 is assigned as DP rank 0, PP rank 0, PCP rank 0, "
|
|
222
|
+
"TP rank 0, EP rank N/A, EPLB rank N/A\n")
|
|
223
|
+
f = servediff.parse(text)["facts"]
|
|
224
|
+
assert f["parallel.world_size"].value == 1
|
|
225
|
+
assert "ETP" not in f["parallel.roles"].value # 0.28 format lacks the field
|
|
226
|
+
assert "kv_cache.layout" not in f
|
|
@@ -58,3 +58,15 @@ def test_env_command_writes_file(tmp_path):
|
|
|
58
58
|
|
|
59
59
|
def test_env_command_reports_bad_interpreter(tmp_path):
|
|
60
60
|
assert cli.main(["env", "--python", str(tmp_path / "no-such-python")]) == 2
|
|
61
|
+
|
|
62
|
+
|
|
63
|
+
def test_env_diff_markdown_output(tmp_path, capsys):
|
|
64
|
+
a = tmp_path / "a.json"; b = tmp_path / "b.json"
|
|
65
|
+
a.write_text(json.dumps(snap({"torch": "2.11.0", "vllm": "0.28.0"})))
|
|
66
|
+
b.write_text(json.dumps(snap({"torch": "2.11.0", "vllm": "0.30.0"})))
|
|
67
|
+
assert cli.main(["diff", str(a), str(b), "--markdown"]) == 1
|
|
68
|
+
out = capsys.readouterr().out
|
|
69
|
+
assert out.startswith("# servediff report")
|
|
70
|
+
assert "| `pkg.vllm` | 0.28.0 | 0.30.0 | changed |" in out
|
|
71
|
+
# unchanged torch row is hidden by default
|
|
72
|
+
assert "pkg.torch" not in out
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|
|
File without changes
|